Create the InferCNV Object

Reading in the raw counts matrix and meta data, populating the infercnv object

infercnv_obj = CreateInfercnvObject(
  raw_counts_matrix="oligodendroglioma_expression_downsampled.counts.matrix",
  annotations_file="oligodendroglioma_annotations_downsampled.txt",
  delim="\t",
  gene_order_file="gencode_downsampled.txt",
  ref_group_names=c("Microglia/Macrophage","Oligodendrocytes (non-malignant)"))
## INFO [2018-11-07 15:11:59] ::order_reduce:Start.
## INFO [2018-11-07 15:11:59] .order_reduce(): expr and order match.
## INFO [2018-11-07 15:11:59] ::process_data:order_reduce:Reduction from positional data, new dimensions (r,c) = 10338,184 Total=18322440.6799817 Min=0 Max=34215.
## INFO [2018-11-07 15:11:59] validating infercnv_obj

Filtering genes

Removing those genes that are very lowly expressed or present in very few cells

# filter out low expressed genes
cutoff=1
infercnv_obj <- require_above_min_mean_expr_cutoff(infercnv_obj, cutoff)
## INFO [2018-11-07 15:11:59] ::above_min_mean_expr_cutoff:Start
## INFO [2018-11-07 15:11:59] Removing 1510 genes from matrix as below mean expr threshold: 1
## INFO [2018-11-07 15:11:59] validating infercnv_obj
## INFO [2018-11-07 15:11:59] There are 8828 genes and 184 cells remaining in the expr matrix.
# filter out bad cells
min_cells_per_gene=3
infercnv_obj <- require_above_min_cells_ref(infercnv_obj, min_cells_per_gene=min_cells_per_gene)
## INFO [2018-11-07 15:11:59] no genes removed due to min cells/gene filter
## for safe keeping
infercnv_orig_filtered = infercnv_obj
#plot_mean_chr_expr_lineplot(infercnv_obj)
save('infercnv_obj', file = 'infercnv_obj.orig_filtered')

Normalize each cell’s counts for sequencing depth

Perform a total sum normalization. Generates counts-per-million or counts-per-100k, depending on the overall sequencing depth.

infercnv_obj <- infercnv:::normalize_counts_by_seq_depth(infercnv_obj)
## INFO [2018-11-07 15:12:00] Computed total sum normalization factor as: 100000.000000

Spike in artificial variation for tracking purposes

Add ~0x and 2x variation to an artificial spike-in data set based on the normal cells so we can track and later scale residual expression data to this level of variation.

infercnv_obj <- spike_in_variation_chrs(infercnv_obj)
## INFO [2018-11-07 15:12:00] Selecting longest chrs for adding spike: chr1,chr2
## INFO [2018-11-07 15:12:00] processing group: malignant_MGH36
## INFO [2018-11-07 15:12:01] processing group: malignant_MGH53
## INFO [2018-11-07 15:12:01] processing group: malignant_93
## INFO [2018-11-07 15:12:01] processing group: malignant_97
## INFO [2018-11-07 15:12:02] processing group: Microglia/Macrophage
## INFO [2018-11-07 15:12:02] processing group: Oligodendrocytes (non-malignant)
## INFO [2018-11-07 15:12:02] processing group: malignant_MGH36
## INFO [2018-11-07 15:12:02] processing group: malignant_MGH53
## INFO [2018-11-07 15:12:02] processing group: malignant_93
## INFO [2018-11-07 15:12:02] processing group: malignant_97
## INFO [2018-11-07 15:12:02] processing group: Microglia/Macrophage
## INFO [2018-11-07 15:12:03] processing group: Oligodendrocytes (non-malignant)
## INFO [2018-11-07 15:12:10] validating infercnv_obj

perform Anscombe normalization

<<<<<<< HEAD Suggested for removing noisy variation at low counts ======= Useful noise reduction method.
See: https://en.wikipedia.org/wiki/Anscombe_transform >>>>>>> 29a0b973d2701fe5ea2834efcd6a82dd542e0308

infercnv_obj <- infercnv:::anscombe_transform(infercnv_obj)
save('infercnv_obj', file='infercnv_obj.anscombe')

log transform the normalized counts:

infercnv_obj <- log2xplus1(infercnv_obj)
## INFO [2018-11-07 15:12:11] transforming log2xplus1()
save('infercnv_obj', file='infercnv_obj.log_transformed')

Apply maximum bounds to the expression data to reduce outlier effects

Here we define a threshold by taking the mean of the bounds of expression data across all cells. This is then use to define a cap for the bounds of all data.

threshold = mean(abs(get_average_bounds(infercnv_obj))) 
infercnv_obj <- apply_max_threshold_bounds(infercnv_obj, threshold=threshold)
## INFO [2018-11-07 15:12:13] ::process_data:setting max centered expr, threshold set to: +/-:  4.22562639983676

Initial view, before inferCNV operations:

plot_cnv(infercnv_obj, 
         output_filename='infercnv.logtransf', 
         x.range="auto", 
         title = "Before InferCNV (filtered & log2 transformed)", 
         color_safe_pal = FALSE, 
         x.center = mean(infercnv_obj@expr.data))
knitr::include_graphics("infercnv.logtransf.png")

perform smoothing across chromosomes

The expression values are

infercnv_obj = smooth_by_chromosome(infercnv_obj, window_length=101, smooth_ends=TRUE)
## INFO [2018-11-07 15:12:46] ::smooth_window:Start.
## INFO [2018-11-07 15:12:46] ::smooth_window:Start.
## INFO [2018-11-07 15:12:46] ::smooth_window:Start.
## INFO [2018-11-07 15:12:46] ::smooth_window:Start.
## INFO [2018-11-07 15:12:47] ::smooth_window:Start.
## INFO [2018-11-07 15:12:47] ::smooth_window:Start.
## INFO [2018-11-07 15:12:47] ::smooth_window:Start.
## INFO [2018-11-07 15:12:48] ::smooth_window:Start.
## INFO [2018-11-07 15:12:48] ::smooth_window:Start.
## INFO [2018-11-07 15:12:48] ::smooth_window:Start.
## INFO [2018-11-07 15:12:49] ::smooth_window:Start.
## INFO [2018-11-07 15:12:49] ::smooth_window:Start.
## INFO [2018-11-07 15:12:49] ::smooth_window:Start.
## INFO [2018-11-07 15:12:50] ::smooth_window:Start.
## INFO [2018-11-07 15:12:50] ::smooth_window:Start.
## INFO [2018-11-07 15:12:50] ::smooth_window:Start.
## INFO [2018-11-07 15:12:51] ::smooth_window:Start.
## INFO [2018-11-07 15:12:51] ::smooth_window:Start.
## INFO [2018-11-07 15:12:51] ::smooth_window:Start.
## INFO [2018-11-07 15:12:52] ::smooth_window:Start.
## INFO [2018-11-07 15:12:52] ::smooth_window:Start.
## INFO [2018-11-07 15:12:52] ::smooth_window:Start.
## INFO [2018-11-07 15:12:52] ::smooth_window:Start.
## INFO [2018-11-07 15:12:53] ::smooth_window:Start.
save('infercnv_obj', file='infercnv_obj.smooth_by_chr')

# re-center each cell
infercnv_obj <- center_cell_expr_across_chromosome(infercnv_obj, method = "median")
## INFO [2018-11-07 15:12:55] ::center_smooth across chromosomes per cell
save('infercnv_obj', file='infercnv_obj.cells_recentered')
plot_cnv(infercnv_obj, 
         output_filename='infercnv.chr_smoothed', 
         x.range="auto", 
         title = "chr smoothed and cells re-centered", 
         color_safe_pal = FALSE)
knitr::include_graphics("infercnv.chr_smoothed.png")

subtract the reference values from observations, now have log(fold change) values

infercnv_obj <- subtract_ref_expr_from_obs(infercnv_obj, inv_log=TRUE)
## INFO [2018-11-07 15:13:28] ::subtract_ref_expr_from_obs:Start
## INFO [2018-11-07 15:13:29] subtracting mean(normal) per gene per cell across all data
save('infercnv_obj', file='infercnv_obj.ref_subtracted')
plot_cnv(infercnv_obj, 
         output_filename='infercnv.ref_subtracted', 
         x.range="auto", 
         title="ref subtracted", 
         color_safe_pal = FALSE)
knitr::include_graphics("infercnv.ref_subtracted.png")

invert log values

Converting the log(FC) values to regular fold change values, centered at 1 (no fold change)

This is important because we want (1/2)x to be symmetrical to 1.5x, representing loss/gain of one chromosome region.

infercnv_obj <- invert_log2(infercnv_obj)
## INFO [2018-11-07 15:14:00] invert_log2(), computing 2^x
save('infercnv_obj', file='infercnv_obj.inverted_log')
plot_cnv(infercnv_obj, 
         output_filename='infercnv.inverted', 
         color_safe_pal = FALSE, 
         x.range="auto", 
         x.center=1, 
         title = "inverted log FC to FC")
knitr::include_graphics("infercnv.inverted.png")

Removing noise

infercnv_obj <- clear_noise_via_ref_mean_sd(infercnv_obj, sd_amplifier = 1.0)
## INFO [2018-11-07 15:14:31] :: **** clear_noise_via_ref_quantiles **** : removing noise between bounds:  0.950072208332557 - 1.05111441688787
save('infercnv_obj', file='infercnv_obj.denoised')
plot_cnv(infercnv_obj, 
         output_filename='infercnv.denoised', 
         x.range="auto", 
         x.center=1, 
         title="denoised", 
         color_safe_pal = FALSE)
knitr::include_graphics("infercnv.denoised.png")

Remove outlier data points

This generally improves on the visualization

infercnv_obj = remove_outliers_norm(infercnv_obj)
## INFO [2018-11-07 15:15:01] ::remove_outlier_norm:Start out_method: average_bound lower_bound: NA upper_bound: NA
## INFO [2018-11-07 15:15:01] ::remove_outlier_norm:Start out_method: average_bound lower_bound: NA upper_bound: NA
## INFO [2018-11-07 15:15:01] ::remove_outlier_norm using method: average_bound for defining outliers.
## INFO [2018-11-07 15:15:01] outlier bounds defined between: 0.655407 - 1.36905
save('infercnv_obj', file="infercnv_obj.outliers_removed")
plot_cnv(infercnv_obj, 
         output_filename='infercnv.outliers_removed', 
         color_safe_pal = FALSE, 
         x.range="auto", 
         x.center=1, 
         title = "outliers removed")
knitr::include_graphics("infercnv.outliers_removed.png")

Scale residual expression values according to the Spike-in

Perform rescaling of the data according to the spike-in w/ preset variation levels. Then, remove the spike-in data.

# rescale
infercnv_obj <- scale_cnv_by_spike(infercnv_obj)
# remove the spike-in
infercnv_obj <- remove_spike(infercnv_obj)
## INFO [2018-11-07 15:15:36] Removing spike

Mask out those genes that are not signficantly different from the normal cells

Runs a Wilcoxon rank test comparing tumor/normal for each patient and normal sample, and masks out those genes that are not significantly DE.

infercnv_obj <- infercnv:::mask_non_DE_genes_basic(infercnv_obj, test.use = ‘t’, center_val=1)

save(‘infercnv_obj’, file=“infercnv_obj.non_DE_masked”)

## INFO [2018-11-07 15:15:36] Finding DE genes between malignant_MGH36 and Microglia/Macrophage
## INFO [2018-11-07 15:15:40] Found 5007 genes / 8828 total as DE
## INFO [2018-11-07 15:15:40] Finding DE genes between malignant_MGH36 and Oligodendrocytes (non-malignant)
## INFO [2018-11-07 15:15:44] Found 4271 genes / 8828 total as DE
## INFO [2018-11-07 15:15:44] Finding DE genes between malignant_MGH53 and Microglia/Macrophage
## INFO [2018-11-07 15:15:48] Found 3945 genes / 8828 total as DE
## INFO [2018-11-07 15:15:48] Finding DE genes between malignant_MGH53 and Oligodendrocytes (non-malignant)
## INFO [2018-11-07 15:15:52] Found 3194 genes / 8828 total as DE
## INFO [2018-11-07 15:15:52] Finding DE genes between malignant_93 and Microglia/Macrophage
## INFO [2018-11-07 15:15:56] Found 4651 genes / 8828 total as DE
## INFO [2018-11-07 15:15:56] Finding DE genes between malignant_93 and Oligodendrocytes (non-malignant)
## INFO [2018-11-07 15:16:00] Found 3889 genes / 8828 total as DE
## INFO [2018-11-07 15:16:00] Finding DE genes between malignant_97 and Microglia/Macrophage
## INFO [2018-11-07 15:16:04] Found 4823 genes / 8828 total as DE
## INFO [2018-11-07 15:16:04] Finding DE genes between malignant_97 and Oligodendrocytes (non-malignant)
## INFO [2018-11-07 15:16:08] Found 4470 genes / 8828 total as DE
plot_cnv(infercnv_obj, 
         output_filename='infercnv.non-DE-genes-masked', 
         color_safe_pal = FALSE, 
         x.range=c(0,2), # want 0-2 post scaling by the spike-in 
         x.center=1, 
         title = "non-DE-genes-masked")
knitr::include_graphics("infercnv.non-DE-genes-masked.png")

Brighten it up by changing the scale threshold to our liking:

plot_cnv(infercnv_obj, 
         output_filename='infercnv.finalized_view', 
         color_safe_pal = FALSE, 
         x.range=c(0.7, 1.3), 
         x.center=1, 
         title = "InferCNV")
## INFO [2018-11-07 15:16:29] ::plot_cnv:Start
## INFO [2018-11-07 15:16:29] ::plot_cnv:Current data dimensions (r,c)=8828,184 Total=1616042.7561869 Min=0 Max=2.05024441345348.
## INFO [2018-11-07 15:16:29] ::plot_cnv:Depending on the size of the matrix this may take a moment.
## INFO [2018-11-07 15:16:29] plot_cnv_observation:Start
## INFO [2018-11-07 15:16:29] Observation data size: Cells= 142 Genes= 8828
## INFO [2018-11-07 15:16:29] clustering observations via method: ward.D
## INFO [2018-11-07 15:16:29] Number of genes in group(1) is 33
## INFO [2018-11-07 15:16:29] group size being clustered:  33,8828
## INFO [2018-11-07 15:16:29] Number of genes in group(2) is 34
## INFO [2018-11-07 15:16:29] group size being clustered:  34,8828
## INFO [2018-11-07 15:16:29] Number of genes in group(3) is 40
## INFO [2018-11-07 15:16:29] group size being clustered:  40,8828
## INFO [2018-11-07 15:16:29] Number of genes in group(4) is 35
## INFO [2018-11-07 15:16:29] group size being clustered:  35,8828
## INFO [2018-11-07 15:16:29] plot_cnv_observation:Writing observation groupings/color.
## INFO [2018-11-07 15:16:42] Colors for breaks:  #00008B,#24249B,#4848AB,#6D6DBC,#9191CC,#B6B6DD,#DADAEE,#FFFFFF,#EEDADA,#DDB6B6,#CC9191,#BC6D6D,#AB4848,#9B2424,#8B0000
## INFO [2018-11-07 15:16:42] Quantiles of plotted data range: 0.7,1,1,1.00168846950822,1.3
## INFO [2018-11-07 15:16:42] plot_cnv_references:Writing observation data to ./infercnv.finalized_view.observations.txt
## INFO [2018-11-07 15:16:43] plot_cnv_references:Start
## INFO [2018-11-07 15:16:43] Reference data size: Cells= 42 Genes= 8828
## INFO [2018-11-07 15:16:43] plot_cnv_references:Number reference groups= 2
## INFO [2018-11-07 15:16:43] plot_cnv_references:Plotting heatmap.
## INFO [2018-11-07 15:16:47] Colors for breaks:  #00008B,#24249B,#4848AB,#6D6DBC,#9191CC,#B6B6DD,#DADAEE,#FFFFFF,#EEDADA,#DDB6B6,#CC9191,#BC6D6D,#AB4848,#9B2424,#8B0000
## INFO [2018-11-07 15:16:47] Quantiles of plotted data range: 0.7,1.00168846950822,1.00168846950822,1.00168846950822,1.3
## INFO [2018-11-07 15:16:47] plot_cnv_references:Writing reference data to ./infercnv.finalized_view.references.txt
## quartz_off_screen 
##                 2
knitr::include_graphics("infercnv.finalized_view.png")

And that’s it. You can experiment with each step to fine-tune your data exploration. See the documentation for uploading the resulting data matrix into the Next Generation Clustered Heatmap Viewer for more interactive exploration of the infercnv-processed data: https://github.com/broadinstitute/inferCNV/wiki/Next-Generation-Clustered-Heat-Map